Data Visualization in R 2. Standard graphics in R

Michael Friendly SCS Short Course

Sep/Oct, 2018 http://datavis.ca/courses/RGraphics/

Course outline

1. Overview of R graphics 2. Standard graphics in R 3. Grid & lattice graphics 4. ggplot2

2

Outline: Session 2 • Session 2: Standard graphics in R R object-oriented design

Tweaking graphs: control graphic parameters

• Colors, point symbols, line styles • Labels and titles

Annotating graphs • Add fitted lines, confidence envelopes • Add text, legends, point labels

R graphics systems • Two graphics worlds “graphics”– traditional or

base graphics “grid”– new style graphics

• Things work very differently in these

• Infrastructure for both is “grDevices” – the R graphics engine Graphics devices, colors, fonts

4

e.g., • the Cairo graphics device can create high-quality

vector (PDF, PostScript and SVG) and bitmap output (PNG,JPEG,TIFF)

• the tikz device uses the LaTeX tikz package and LaTeX fonts, colors, etc.

Base graphics functions: high & low • Graphics functions are mostly one of two types: High-level functions → complete plots

• plot(), boxplot(), dotplot(), mosaicplot(), … Low-level functions → add to an existing plot

• lines(), points(), legend(), arrows(), polygon(), text()

Some functions can work either way, via an argument add=TRUE/FALSE • symbols(x, y, …, add=TRUE) • car::dataEllipse(x, y, add=TRUE, …)

5

The many faces of plot() • plot() is the most important function in traditional

graphics • It is designed as a generic function, that does

different things with numeric data (x, y), factors (FAC), matrices (MAT),… plot(x) - index plot of x[i] vs I plot(x, y) – scatterplot plot(FAC, y) – boxplots plot(x, FAC) – stripchart plot(FAC, FAC) – spineplot, barchart plot(MAT) – scatterplot matrix -> pairs()

6

Object-oriented approach in R • Everything in R is an object, and has a class data sets: class “data.frame” statistical models: class “lm”, “glm”, …

• Fit a model: obj <- lm(…) → a “lm” model object print(obj) & summary(obj) → numerical results anova(obj) & Anova(obj) → tests for model terms update(obj), add1(obj), drop1(obj) model selection

Objects & methods

8

Method dispatch: The S3 object system

• Functions return objects of a given class • Anova/regression: lm() → an “lm” object • Generalized linear models: glm() → c(“glm”, “lm”) – also inherits from lm() • Loglinear models: loglm() → a “loglm” object

• Class-specific methods have names of the form method.class • plot.lm(), plot.glm() – model diagnostic plots

• Generic functions– print(), plot(), summary() call the appropriate method for the class

• plot(Effect(obj)) – calls plot.eff() effect plots • plot(influence(obj)) – calls plot.influence() for influence plots • plot(prcomp(obj)) – plots a PCA solution for a “prcomp” object

R objects & methods

9

> data(Duncan, package="car") > class(Duncan) [1] "data.frame" > duncan.mod <- lm(prestige ~ income + education, data=Duncan) > class(duncan.mod) [1] "lm" > print(duncan.mod) Call: lm(formula = prestige ~ income + education, data = Duncan) Coefficients: (Intercept) income education -6.065 0.599 0.546 > Anova(duncan.mod) Anova Table (Type II tests) Response: prestige Sum Sq Df F value Pr(>F) income 4474 1 25.0 1.1e-05 *** education 5516 1 30.9 1.7e-06 *** Residuals 7507 42 --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

→ print.lm()

→ Anova.lm()

Objects & methods

10

> library(car) > methods(class="lm") [1] add1 alias anova Anova [5] avPlot Boot bootCase boxCox [9] case.names ceresPlot coerce confidenceEllipse [13] confint cooks.distance crPlot deltaMethod [17] deviance dfbeta dfbetaPlots dfbetas [21] dfbetasPlots drop1 dummy.coef durbinWatsonTest [25] effects extractAIC family formula [29] hatvalues hccm infIndexPlot influence [33] influencePlot initialize inverseResponsePlot kappa [37] labels leveneTest leveragePlot linearHypothesis [41] logLik mcPlot mmp model.frame [45] model.matrix ncvTest nextBoot nobs [49] outlierTest plot powerTransform predict [53] print proj qqnorm qqPlot [57] qr residualPlot residualPlots residuals [61] rstandard rstudent show sigmaHat [65] simulate slotsFromS3 spreadLevelPlot summary [69] variable.names vcov see '?methods' for accessing help and source code

Some methods for “lm” objects (in the base and car packages):

Plot methods

11

> methods("plot") [1] plot.acf* plot.ACF* plot.augPred* [4] plot.coef.mer* plot.compareFits* plot.correspondence* [7] plot.data.frame* plot.decomposed.ts* plot.default [10] plot.dendrogram* plot.density* plot.ecdf [13] plot.factor* plot.formula* plot.function [16] plot.gam* plot.gls* plot.hclust* [19] plot.histogram* plot.HoltWinters* plot.intervals.lmList* [22] plot.isoreg* plot.jam* plot.lda* [25] plot.lm* plot.lme* plot.lmList* [28] plot.lmList4* plot.lmList4.confint* plot.mca* [31] plot.medpolish* plot.merMod* plot.mlm* [34] plot.nffGroupedData* plot.nfnGroupedData* plot.nls* [37] plot.nmGroupedData* plot.PBmodcomp* plot.pdMat* [40] plot.powerTransform* plot.ppr* plot.prcomp* [43] plot.princomp* plot.profile* plot.profile.nls* [46] plot.qss1* plot.qss2* plot.ranef.lme* [49] plot.ranef.lmList* plot.ranef.mer* plot.raster* [52] plot.ridgelm* plot.rq.process* plot.rqs* [55] plot.rqss* plot.shingle* plot.simulate.lme* [58] plot.skewpowerTransform* plot.spec* plot.spline* [61] plot.stepfun plot.stl* plot.summary.crqs* [64] plot.summary.rqs* plot.summary.rqss* plot.table* [67] plot.table.rq* plot.trellis* plot.ts [70] plot.tskernel* plot.TukeyHSD* plot.Variogram* [73] plot.xyVector* see '?methods' for accessing help and source code

Some available plot() methods

12

op <- par(mfrow=c(1,4)) # change layout parameters plot(duncan.mod) # regression diagnostic plots par(op) # restore old parameters

Some plot methods produce multiple plots. You can control the layout with par() settings

avPlots(duncan.mod, id.n=2, pch=16, ellipse=TRUE, ellipse.args=list(levels=0.68, fill=TRUE, fill.alpha=0.1))

Some plot methods have lots of optional arguments for graphic enhancements. Use help() for documentation

Graphic parameters • All graphic functions take arguments that control

details colors (col=) point symbols (pch=) line type (lty=); line width (lwd=)

• Often these have default values in the function definition col=“black”; col=par(“col”); col=palette()[1] lwd=2, lty=1

• Most high-level graphic functions have a “…” argument that allow passing other arguments to the plotting functions

13

Graphic parameters • Some graphics parameters can be set globally for all

graphs in your session, using the par() function par(mar=c(4,4,1,1)) – plot margins par(cex.lab=1.5) – make axis labels 50% larger par(cex=2) – make text & point symbols 2x larger Graphics functions often use these as defaults

• Most can be set in calls to high-level plotting functions avPlots(duncan.mod, pch=16, cex=2, cex.lab=1.5, …)

14

From: http://gastonsanchez.com/r-graphical-parameters-cheatsheet.pdf

Graphic parameters

16

The most commonly used graphic parameters:

These colors are the default, palette() R packages specifically related to color: colorspace, colorRamps, RColorBrewer, ..

Plot types

17

The functions plot(), points() and lines() understand a type= parameter and render the (x, y) values in different ways.

x <- -5:5 y <- -x^2 + 25 plot(x, y, type=“p”) plot(x, y, type=“l”) plot(x, y, type=“b”) plot(x, y, type=“o”) plot(x, y, type=“h”) plot(x, y, type=“s”)

More on color • Presentation graphs require careful choice of colors Legible if copied in B/W? Visible to those with color deficiency? Mapping categorical or continuous variables to color scale

• R has a variety of ways to specify color color names: see colors() Hex RGB: red = “#FF0000” , blue=“#0000FF” with transparency: #rrggbbaa hsv(): hue, saturation, value (better as perceptual model) colorRamps: rainbow(n)

18

#0000FFA0 #0000FF80

19

library("colorspace") pal < - choose_palette()

See: https://www.nceas.ucsb.edu/~frazier/RSpatialGuides/colorPaletteCheatsheet.pdf

library(“RColorBrewer") display.brewer.all()

Traditional R graphics: mental model • R graphics functions add ink to a canvas – the

“painter’s model” new graphics elements overlay / obscure what is there before only way to move or remove stuff is to re-draw in white (background

color) animated graphs re-do the whole plot in each frame Transparent colors are often useful for filled areas

• Typically, create a graph with a high-level function, then add to it with low-level if desired

• I’ll illustrate by re-constructing two historical graphs 1. Identify the graphical elements: points, lines, text, … 2. Start with a basic plot, then add to it

20

Building a custom graph

21

Custom graphs can be constructed by adding graphical elements (points, lines, text, arrows, etc.) to a basic plot()

> data(Arbuthnot, package=“HistData”) > head(Arbuthnot[,c(1:3,6,7)]) Year Males Females Ratio Total 1 1629 5218 4683 1.114 9.901 2 1630 4858 4457 1.090 9.315 3 1631 4422 4102 1.078 8.524 4 1632 4994 4590 1.088 9.584 5 1633 5158 4839 1.066 9.997 6 1634 5035 4820 1.045 9.855 … … … … … …

Arbuthnot didn’t make a graph. He simply calculated the probability that in 81 years from 1629—1710, the sex ratio would always be > 1 The first significance test!

John Arbuthnot: data on male/female sex ratios:

reference line

figure caption

regression line & loess smooth

points & lines

Follow along • From the course web page, click on the script

arbutnot.R, http://www.datavis.ca/courses/RGraphics/R/arbuthnot.R

• Select all (ctrl+A) and copy (ctrl+C) to the clipboard • In R Studio, open a new R script file (ctrl+shift+N) • Paste the contents (ctrl+V) • Run the lines (ctrl+Enter) to along with me

(You could instead save that file to your lab HOME directory and open it from there.)

Building a custom graph

23

plot(Ratio ~ Year, data=Arbuthnot, pch=16, ylim=c(1, 1.20), cex.lab = 1.3, ylab="Sex Ratio (M/F)")

1. Start with a basic plot of points

Code details: pch: I like filled circles (16=●) for points ylim: allow more vertical space for caption cex.lab: make axis labels larger

Building a custom graph

24

plot(Ratio ~ Year, data=Arbuthnot, pch=16, ylim=c(1, 1.20), cex.lab = 1.3, ylab="Sex Ratio (M/F)") # connect points by lines lines(Ratio ~ Year, data=Arbuthnot, col="gray")

2. Add gray lines

Code details: I could have used type=“b” or type=“o” But I wanted the lines in gray, not black

Building a custom graph

25

plot(Ratio ~ Year, data=Arbuthnot, pch=16, ylim=c(1, 1.20), cex.lab = 1.3, ylab="Sex Ratio (M/F)") # connect points by lines lines(Ratio ~ Year, data=Arbuthnot, col="gray") # add reference line abline(h=1, col="red", lwd=3) text(1640, 1, "Males = Females", col="red")

3. Add horizontal reference line & label

reference line

Building a custom graph

26

plot(Ratio ~ Year, data=Arbuthnot, pch=16, ylim=c(1, 1.20), cex.lab = 1.3, ylab="Sex Ratio (M/F)") # connect points by lines lines(Ratio ~ Year, data=Arbuthnot, col="gray") # add reference line abline(h=1, col="red", lwd=3) text(1640, 1, "Males = Females", col="red") # add linear regression line abline(lm(Ratio ~ Year, data=Arbuthnot), col="darkgreen") # add loess smooth Arb.smooth <- with(Arbuthnot, loess.smooth(Year, Ratio)) lines(Arb.smooth$x, Arb.smooth$y, col="blue", lwd=2)

4. Add regression & smoothed lines

Building a custom graph

27

plot(Ratio ~ Year, data=Arbuthnot, pch=16, ylim=c(1, 1.20), cex.lab = 1.3, ylab="Sex Ratio (M/F)") # connect points by lines lines(Ratio ~ Year, data=Arbuthnot, col="gray") # add reference line abline(h=1, col="red", lwd=3) text(1640, 1, "Males = Females", col="red") # add linear regression line abline(lm(Ratio ~ Year, data=Arbuthnot), col="darkgreen") # add loess smooth Arb.smooth <- with(Arbuthnot, loess.smooth(Year, Ratio)) lines(Arb.smooth$x, Arb.smooth$y, col="blue", lwd=2) # add internal figure caption text(1690, 1.19, "Arbuthnot's data on the\nMale / Female Sex Ratio", cex=1.2)

5. Add figure caption figure caption

28

library(ggplot2) ggplot(Arbuthnot, aes(x=Year, y=Ratio)) + ylim(1, 1.20) + ylab("Sex Ratio (M/F)") + geom_point(pch=16, size=2) + geom_line(color="gray") + geom_smooth(method="loess", color="blue", fill="blue", alpha=0.2) + geom_smooth(method="lm", color="darkgreen", se=FALSE) + geom_hline(yintercept=1, color="red", size=2) + annotate("text", x=1640, y=1.005, label="Males = Females", color="red", size=4) + annotate("text", x=1690, y=1.19, label="Arbuthnot's data on the\nMale / Female Sex Ratio", size=6) + theme_bw()

The same graph, using ggplot2 ggplot2 has a totally different idea about constructing graphs The syntax adds elements and layers to a graph with functions connected with “+” signs. Details in a following lecture

ggplot code:

Playfair’s wheat

29

William Playfair (1759—1836) invented most of the forms of modern data graphics: the bar chart, line graph and pie chart. • This multivariate chart shows the price of wheat (bars), wages of a good mechanic

(line graph), and the reigns of British monarchs over 250 years, 1565—1830 • Playfair’s goal: Show that workers were better off now than at any time in the past.

Did Playfair succeed? What can you read from this chart re: wages vs. price of wheat?

Reproducing Playfair’s chart

30

To try to reproduce this chart: • Identify the graphical elements: 3 time series, cartouche caption, grid lines, … • Make a basic plot setting up (x,y) range, axis labels, … • Use low-level functions to add graphical elements

Time series of wheat: plot as step function: type=“s”

Wages: draw using lines()

Caption: plot using multiline text()

Monarchs: draw using segments(), label with text()

Reproducing Playfair’s chart

31

Playfair’s data was digitized from his chart. The HistData package records this as two data frames.

> str(Wheat) 'data.frame': 53 obs. of 3 variables: $ Year : int 1565 1570 1575 1580 1585 1590 1595 1600 1605 1610 ... $ Wheat: num 41 45 42 49 41.5 47 64 27 33 32 ... $ Wages: num 5 5.05 5.08 5.12 5.15 5.25 5.54 5.61 5.69 5.78 ... > str(Wheat.monarchs) 'data.frame': 12 obs. of 4 variables: $ name : Factor w/ 12 levels "Anne","Charles I",..: 5 10 2 4 3 11 12 1 6 7 ... $ start : int 1565 1603 1625 1649 1660 1685 1689 1702 1714 1727 ... $ end : int 1603 1625 1649 1660 1685 1689 1702 1714 1727 1760 ... $ commonwealth: int 0 0 0 1 0 0 0 0 0 0 ...

Code for this example: http://datavis.ca/courses/RGraphics/R/playfair-wheat.R

Reproducing Playfair’s chart

32

with(Wheat, { plot(Year, Wheat, type="s", ylim=c(0,105), ylab="Price of the Quarter of Wheat (shillings)", panel.first=grid(col=gray(.9), lty=1)) lines(Year, Wages, lwd=3, col="red") })

The basic plot is a step-curve for wheat (type=“s”) Add lines for Wages

The area beneath the curve could be filled, using polygon()

with( Wheat, { expressions } ) makes the variables in Wheat available in evaluating the {expressions} I could have used: plot(Wheat ~ Year, data=Wheat, …)

Reproducing Playfair’s chart

33

# label the curve of Wages text(1625,10, "Weekly wages of a good mechanic", cex=0.8, srt=3, col="red") # cartouche text(1650, 85, "Chart", cex=2, font=2) text(1650, 70, paste("Shewing at One View", "The Price of the Quarter of Wheat", "& Wages of Labor by the Week", "from the Year 1565 to 1821", "by William Playfair", sep="\n"), font=3)

text label: srt=3 rotates the text 3o

paste(s1, s2, …, sep=“\n”) makes separate lines font=3: italic

The decorative cartouche is drawn with text(), using a vector of strings

Reproducing Playfair’s chart

34

with(Wheat.monarchs, { y <- ifelse( !commonwealth & (!seq_along(start) %% 2), 102, 104) segments(start, y, end, y, col="black", lwd=7, lend=1) segments(start, y, end, y, col=ifelse(commonwealth, "white", NA), lwd=4, lend=1) text((start+end)/2, y-2, name, cex=0.5) })

The timeline for monarchs is drawn using segments() This part is tricky because I need to calculate the y value for each segment and position the labels accordingly.

alternate y values as (102, 104)

re-draw the white one

Consulting for Playfair

35

WP: Can you help me make a better graph? SCS: Yes, plot the ratio of Wheat / Wages : the labor cost to buy a quarter of wheat

This clearly shows that wheat was becoming cheaper in terms of the amount of labor required Plotting data was so new that Playfair did not think of plotting a derived value.

Consulting for Playfair

36

Wheat1 <- within(na.omit(Wheat), {Labor=Wheat/Wages}) with(Wheat1, { plot(Year, Labor, type='b', pch=16, cex=1.5, lwd=1.5, ylab="Labor cost of a Quarter of Wheat (weeks)", ylim=c(1,12.5), xlim=c(1560,1823), cex.axis=1.2, cex.lab=1.5, lab=c(12,5,7) ); lines(lowess(Year, Labor), col="red", lwd=3) })

The remainder of the code is similar to that for the original plot

Code for this plot: calculate ratio

Galton’s peas

37

In 1875 Francis Galton studied heredity of physical traits. In one experiment, he sent packets of sweet peas of 7 different sizes to friends, and measured the sizes of their offspring. His first attempt was a semi-graphic table, tabulating the number of parent-child seeds in each combination of values. He noted that both distributions followed the “law of frequency of error” (Normal distribution)

Galton’s peas: The first regression line

38

Galton’s (1877) presentation graph: • Plotted mean diameter of child seeds vs. mean of parents • Noticed these were nearly in a line– An “Ah ha” moment! • The slope of the line said something about heredity

But, the slope of the line < 1 → “reversion” toward mean → children of large/small parents less extreme than their parents Later used the term “regression” for this phenomenon, and statistical explanation

Image: From K. Pearson, The Life, Letters and Labours of Francis Galton, Volume 3A, Chapter 14, Fig. 1

Galton’s peas: Plotting discrete data

39

How Galton got there – the untold story His friend, JFW Herschel said, “Why don’t you make a scatterplot?” He fired up R on his Babbage machine … … but was initially disappointed in the result: too much overplotting

data(peas, package="psych") plot(child ~ parent, data=peas, pch=16, cex.lab=1.25, asp=1, xlim=c(14, 23), xlab="Parent seed diameter (0.01 in.)", ylab="Child seed diameter (0.01 in.)")

NB: Galton was careful to • Set aspect ratio = 1 • Use explicit axis labels

Another great R historical moment

Code: http://datavis.ca/courses/RGraphics/R/galton-peas.R

Galton’s peas: Data wrangling

40

Galton thoughtfully met with an SCS consultant, who said: “Show me your data!!!”

> str(peas) 'data.frame': 700 obs. of 2 variables: $ parent: num 21 21 21 21 21 21 21 21 21 21 ... $ child : num 14.7 14.7 14.7 14.7 14.7 ...

SCS: Summarize them with dplyr

library(dplyr) peas.freq <- peas %>% group_by(parent, child) %>% summarise( count=n() )

> peas.freq Source: local data frame [52 x 3] Groups: parent [?] parent child count <dbl> <dbl> <int> 1 15 13.77 46 2 15 14.77 14 3 15 15.77 9 4 15 16.77 11 5 15 17.77 14 6 15 18.77 4 7 15 19.77 2 8 16 14.28 34 9 16 15.28 15 10 16 16.28 18 # ... with 42 more rows

SCS: Ah! your data are discrete.

Galton’s peas: a text-table plot

41

plot(child ~ parent, data=peas, …) with( peas.freq, text( parent, child, count, col="red", cex=log(count)))

Galton: Ah! Maybe I’ll just go back to my original table SCS consultant: Good, but make it into a plot also: use text() Here’s a good graphic trick: make font size ~ f(n)

size ~ log(n)

text label

Galton’s peas: Sunflower plots

42

sunflowerplot(child ~ parent, data=peas, pch=16, cex.lab=1.25, asp=1, xlim=c(14, 23), xlab="Parent seed diameter (0.01 in.)", ylab="Child seed diameter (0.01 in.)")

Perhaps better: use point symbols that show explicitly the number of observations at each (x, y) location A sunflower plot uses symbols with the number of rays = # of obs at each (x, y) Now, he could see the upward trend – sort of

Galton’s peas: jittering

43

plot(jitter(child) ~ jitter(parent), data=peas, pch=16, cex.lab=1.25, asp=1, xlim=c(14, 23), xlab="Parent seed diameter (0.01 in.)", ylab="Child seed diameter (0.01 in.)")

Another possibility is to jitter() the plotted points by adding little random #s But, he also needed to calculate and plot the line of means and the trend line

# add line of means means <- aggregate(child ~ parent, data=peas, FUN=mean) lines (child ~ parent, data=means, type="b", pch="+", cex=2, lwd=7, col="darkgreen") text(15, 15.3, "means", col="darkgreen", cex=1.4, pos=2) # calculate & draw the regression line peas.mod <- lm(child ~ parent, data=peas) abline(peas.mod, lwd=2, col=“blue”)

Plotting discrete data: Galton’s peas

44

Making Galton’s argument visually clearer: • Label the regression line with its slope • Show the comparison line (slope=1) if there was no regression toward the mean

text(23, y=18.3, "child ~0.34 * parent", cex=1.4, col="blue") # line of unit slope mx <- mean(peas$parent) my <- mean(peas$child) xp <- 14:22 yp <- my + 1* (xp - mx) lines(xp, yp, col="darkgray", lwd=3, lty="longdash") text(23.2, yp[9], "child ~ 1 * parent", cex=1.4, col="darkgray")

Just for fun: CatterPlots

45

library(devtools) install_github("Gibbsdavidl/CatterPlots") # plot random cats multicat(xs=-10:10, ys=rnorm(21), cat=c(1,2,3,4,5,6,7,8,9,10), catcolor=list(c(0,0,0,1)), canvas=c(-0.1,1.1, -0.1, 1.1), xlab="some cats", ylab="other cats", main="Random Cats")

How this works: • 11 cat shape images saved as PNG • Calls plot(x, y, …) – set up plot frame • rasterImage(catImg, …) – plot each cat

Time series plots

46

R has special methods for dealing with time series data The sunspots data set records monthly mean relative sunspot numbers from 1749 to 1983

> data(sunspots) > str(sunspots) Time-Series [1:2820] from 1749 to 1984: 58 62.6 70 55.7 85 83.5 94.8 75.5 ... plot(sunspots, cex.lab=1.5, ylab="Number of sunspots")

But the aspect ratio (V/H) of the plot is often important. A systematic pattern is revealed when the average local trend is ~ 45o

Time series: lag plots

47

Lag plots show a time series against lagged versions of themselves. This helps visualizing ‘auto-dependence’.

plot(sunspots, lag(sunspots, 1), cex=0.7, col="blue")

There is a strong dependence between this year’s sunspots and last.

Time series: lag plots

48

Often, we want to see dependence across a range of lag values. lag.plot(series) does this quite flexibly

lag.plot(sqrt(sunspots), set = c(1:4, 9:12), layout=c(2,4), col="blue", cex=0.7 )

dependence is persistent, but weakens over lags

Time series: Seasonal patterns

49

Data UKLungDeaths: monthly deaths from bronchitis, emphysema and asthma in the UK, 1974–1979

data(UKLungDeaths) plot(ldeaths, lwd=2, main="UK Lung Deaths")

Looks like cycles, peaking in winter

Time series: Seasonal patterns

50

The acf() function calculates and plots autocorrelations of a time series across various lags

acf(ldeaths, lwd=5, main="Autocorrelations of UK Lung Deaths")

This gives a compact view of the seasonal pattern Other time-series graphs can show other details

Saving image files: R Studio

51

From the R Studio Plots tab, you can save any image in a variety of types

Some options are available in the menu to control the details of size, shape & image format

For publication purposes, you will often want more control: plot margins, font sizes, figure shape, etc.

Saving image files: R scripts • The default graphics device in R is your computer screen. • In an R script, there are 3 steps:

1. Open a graphics device, with desired parameters • Call png(), jpg(), pdf(), …

2. Create the plot 3. Close the graphics device: dev.off()

52

png(file="bowling.png", width=400, height=400) x <- 1:10 y <- (x - 5)^2 plot(y ~ x, type="b", xlab="Frame", ylab="Bowling score“) dev.off()

1

2

3

Much easier with ggplot2: ggsave()

Saving image files: Margins, fonts

53

png(file="bowling.png", width=400, height=400) op <- par(mar=c(5, 5, 2, 1)) x <- 1:10 y <- (x - 5)^2 plot(y ~ x, type="b", pch=15, cex=2, cex.lab=1.5, xlab="Frame", ylab="Bowling score", main="Why do I have a slump?") par(op) dev.off()

png(file="bowling0.png", width=400, height=400) x <- 1:10 y <- (x - 5)^2 plot(y ~ x, type="b", xlab="Frame", ylab="Bowling score“) dev.off()

Set plot margins, font size for points & labels

Saving image files: R markdown • In R markdown files, use chunk options to control

figures & other output global options -- control all chunks: knitr::opts_chunk$set() individual chunk options

54

```{r setup, include=FALSE, message=FALSE} opts_chunk$set(fig.path="figs/", dev=c("png","pdf"), # devices for figs fig.width=6, fig.height=5, # fig size fig.align="center", dpi=300, # make high res. digits=4, … ) # printed output ```

```{r wheat1, fig.width=9, fig.height=4} data(Wheat) plot(Wheat ~ Year, data=Wheat, type=“s”) … ```

Set global options: Change size for this figure:

Details: see https://yihui.name/knitr/options/

Summary • Standard R graphics High-level (plot()) vs. low level (lines()) functions Understand object-oriented methods

• Graphics parameters Understand the basic ones: col, pch, lty, lwd Use help(par) or cheat sheet to find others For a high-level function, use help(fun)

• Building graphs Think about graphic elements: points, lines, areas, … How these should be rendered: graphical attributes

55